proceed acm sigkdd workshop visual analyt knowledg discoveri integr autom analysi interact explor dwai group dwai topic model abstract tackl problem user document collabor filter gener user group user group benefici rate predict document observ rate appli document case user shown earlier user document gener introduc probabilist model task task find group trivial combinatori problem make computation difficult suggest approxim model urp model group user group document predict combin product expert model combin model achiev predict perform origin model don random strategi iter data mine wide varieti data mine method gener exploratori data analysi method dataset lead problem result found method reflect phenomenon shown result method result depict sens unrel properti data cluster give indic clear cluster structur comput correl variabl show signific correl data case correl determin cluster structur paper problem random data previous discov pattern model account random method iter data mine step data mine process random produc random sampl set data matric satisfi discov pattern model data set statist cluster center occurr count data random method sampl data set similar valu statist origin data set metropoli sampl base local swap achiev describ experi real data demonstr us approach result case result cluster impli result frequent pattern discoveri bayesian solut label switch problem abstract label switch problem unidentifi permut cluster gener latent variabl make interpret result comput mcmc sampl difficult introduc fulli bayesian treatment permut perform altern method comput summari posterior sampl nonparametr bayesian method good solut exist approxim case result promis thesummariesareintuitivelyapp summar cluster defin set point likelihood cluster maxim don random strategi iter data mine wide varieti data mine method gener exploratori data analysi method dataset lead problem result found method reflect phenomenon shown result method result depict sens unrel properti data cluster give indic clear cluster structur comput correl variabl show signific correl data case correl determin cluster structur paper problem random data previous discov pattern model account random method iter data mine step data mine process random produc random sampl set data matric satisfi discov pattern model data set statist cluster center occurr count data random method sampl data set similar valu statist origin data set metropoli sampl base local swap achiev describ experi real data demonstr us approach result case result cluster impli result frequent pattern discoveri random techniqu graph mine graph data activ research area data mine method algorithm propos identifi structur graph evalu result lack framework statist hypothesi test focu paper random techniqu unweight undirect graph random import approach assess statist signific data mine result input graph random method sampl data class graph share structur properti input graph describ altern algorithm base local edg swap metropoli sampl test framework graph data set mine algorithm applic graph cluster frequent subgraph mine random approxim algorithm comput bucket order www elsevi locat ipl random approxim algorithm comput bucket order latent group model user prefer predict introduc latent group model predict relev document user model assum latent group structur user document ey reveal interest implicit queri gaze pattern abstract studi research problem implicit inform retriev queri infer ey movement measur user read retriev document train phase user interest learn map user term role term implicit queri assum map univers queri domain construct queri topic learn data construct control experiment set show system prior inform user search ey movement significantli search case proactiv search instanc system monitor read behaviour user topic contrast search read session set inspect document bias relev stronger strategi search content wise similar document ey movement introduct special issu visual analyt knowledg discoveri ey reveal interest implicit queri gaze pattern abstract studi research problem implicit inform retriev queri infer ey movement measur user read retriev document train phase user interest learn map user term role term implicit queri assum map univers queri domain construct queri topic learn data construct control experiment set show system prior inform user search ey movement significantli search case proactiv search instanc system monitor read behaviour user topic contrast search read session set inspect document bias relev stronger strategi search content wise similar document ey movement learn learn implicit queri gaze pattern absenc explicit queri altern infer user interest implicit feedback signal clickstream ey track interest formul implicit queri search formul task probabilist model interpret kind transfer meta learn probabilist model demonstr outperform earlier kernel base method small scale inform retriev task approxim ratio biclust problem biclust consist simultan cluster row column matrix submatric induc pair row column cluster uniform paper approxim optim biclust appli cluster algorithm independ row column input matrix show solut yield worst case approxim ratio norm valu matric norm real valu matric keyword approxim algorithm biclust cluster approxim ratio biclust problem biclust consist simultan cluster row column matrix submatric induc pair row column cluster uniform paper approxim optim biclust appli cluster algorithm independ row column input matrix show solut yield worst case approxim ratio norm valu matric norm real valu matric keyword approxim algorithm biclust cluster discrimin joint densiti model studi discrimin joint densiti model gener model joint densiti learn maxim discrimin cost function condit likelihood framework deriv gener model gener linear model includ logist regress linear discrimin analysi discrimin mixtur unigram benefit deriv discrimin model joint densiti model easi extend model interpret result miss data treat justifi standard method algorithm discov bucket order data order rank item type import task applic queri process scientif data mine total order item mislead group item practic equal rank bucket order total order ti captur essenti order inform overfit data form concept class total order arbitrari partial order address question find bucket order set item pairwis preced inform item discuss method comput pairwis preced data describ simpl effici algorithm find good bucket order algorithm provabl approxim guarante scale larg dataset provid experiment result artifici real data show us bucket order demonstr accuraci effici algorithm discrimin joint densiti model studi discrimin joint densiti model gener model joint densiti learn maxim discrimin cost function condit likelihood framework deriv gener model gener linear model includ logist regress linear discrimin analysi discrimin mixtur unigram benefit deriv discrimin model joint densiti model easi extend model interpret result miss data treat justifi standard method expect maxim algorithm condit likelihood introduc expect maximizationtyp em algorithm maximum likelihood optim condit densiti applic hidden variabl model distribut exponenti famili algorithm altern view automat step size select gradient ascent amount comput trade guarante step increas likelihood tradeoff make algorithm computation feasibl earlier condit em method theoret basi extend baum welch algorithm discrimin hidden markov model speech recognit compar favour current method experi combin ey movement collabor filter proactiv inform retriev studi task proactiv inform retriev combin implicit relev feedback collabor filter construct control experiment set prototyp applic user find interest scientif articl brows titl implicit feedback infer ey movement signal discrimin hidden markov model estim exist data explicit relev feedback collabor filter carri user rate profil model state art probabilist latent variabl model comput markov chain mont carlo techniqu document titl predict accuraci ey movement collabor filter combin significantli chanc predict accuraci leav room improv show proactiv inform retriev combin sourc relev feedback feasibl dwai latent group model user prefer predict introduc latent group model predict relev document user model assum latent group structur user document 